OSC Landbase Movement Capture to Wekinator

What happened, in order#

A phone went into a pocket at the belly and a person walked outdoors, on land, and moved. The phone reported nine numbers about that movement — three of acceleration, three of turning, three of which way it was facing — and sent them across a network to a machine that was listening.

The first capture ran 74.98 seconds and carried 376 packets. That is 5.01 samples a second. Nobody chose that number; the phone delivered it.

And then it delivered different ones. Later the same afternoon, from the very same phone, a three-minute body-dance came back at 43 samples a second — eight and a half times denser. Four days later, 60.

Nobody wrote down what was asked for. So the archive holds three different answers to "how fast does this instrument listen," and no way to tell which request produced which.

What the numbers are#

The first six are close to what the sensor physically measured. The last three are a guess — a fusion filter's best reconstruction, always a little unsure of north. A model that treats all nine as equally trustworthy is making a mistake.

Two months later, the practice became an instrument#

On 2026-08-16 the capture script became three packages anyone can install:

• @miadi/osc — the wire
• @miadi/movement-conductor — one source, many listeners: the machine that learns the sound, the recorder writing the score, and the meters, all hearing the same instant
• @miadi/termux-motion — the phone in your pocket, as a body-worn instrument

The design principle written into the code: recording and performing are one gesture heard by different listeners. There is no separate "now I record" mode. And any recording can perform again — the archive is an instrument, not a shelf.

Then the practice was asked what field it belonged to#

A published package with a stated contract is a claim about the world. So eight research lanes went looking for the fields that claim answers to — 203 sources, 197 verified live — and came back with three things worth saying out loud.

The first: we know what we cannot hear. Five samples a second means a ceiling of about 2.5. What survives is that the body moved and roughly how fast. What is lost is each individual stride, the rhythm of the gait, the heel striking the ground. At sixty samples a second none of that is lost. The ceiling belongs to a take, not to the instrument — and writing that limit into the record instead of smoothing it away turns out to be the most honest thing the instrument does.

The second: the rate is not broken. The phone was asked for ten samples a second and gave five, because Android treats that request as a suggestion. That is expected behaviour, and the repair is one number — write down what was asked for, beside what arrived.

And here is where the research caught itself. The grounding was written as though 5.01 were the rate, because that is the number the first capture carried and the number everyone had quoted since. It is the slowest take in the folder. The correction did not come from a reviewer or an outsider — it came from the archive, from three summary files sitting in sensor_data/ that had been disagreeing with us quietly since June. That is the whole argument for keeping an honest record, demonstrated on the people keeping it.

The third: one of the words in this episode's own title belongs to someone else. Land base is an active term in Canadian law and policy — Additions to Reserve, Treaty Land Entitlement — meaning a First Nation's owned and controlled territory. This episode used it without saying so. Naming that is not a correction to be filed; it is a relation to be acknowledged.

What is still open#

The project's own test of itself has never been run. It goes like this: play the same learned mapping twice for someone who does not know which is which — once driven by the body that taught it, once by a stranger moving at random. If nobody can tell which one belongs to the story, then the song could have been written without the story, and the central claim fails.

The instrument to run that test already exists. It has existed since the scores became replayable. Nobody has asked it the question yet.

Where the rest of this lives#

The full grounding, its sources and its limits: foundations/ in jgwill/Miadi, mirrored into the IAIP repo where the proposal is written. The inquiry woven to this episode: inquiry/ep83-movement-foundations-the-instrument-gets-its-fields-260816-…. The directions the toolset might take, each with its own argument against: rispecs/movement-conductor-family/ in jgwill/Miadi.

This episode continues into Episode 294, the Season 4 room, where the toolset becomes something an audience can meet.

🌸: It started because dancing with a phone was fun. Everything since — the packages, the fields, the honest ceiling, the word we had to give back — grew out of one walk that somebody thought to record.